fak
Agent kernel that routes AI models, caches context, and audits tool calls
Visit anthony-chaudhary.github.io ↗What is fak?
fak is a single Go binary that sits between an AI agent and the tools it calls, acting as a kernel that intercepts and governs all external operations. It works with Claude Code, Cursor, Codex, or any OpenAI/Anthropic client without requiring rewrites. The system reduces costs on long sessions by routing each call to the right model, eliminating wasted turns, and providing auditable verdicts for every tool execution.
Designed for developers running local agents, fak combines native inference, speculative decoding, and agentic caching to accelerate workflows automatically. It enforces capability policies, reuses context across turns, and keeps all decisions inspectable. The maker indicates it was built with Claude Code and Codex.
Momentum
Daily activity over the last 30 days: upvotes, unique visits, new GitHub stars and Hacker News points.
How do we know it's vibe coded?
vibetrend discovered fak automatically. Evidence: GitHub anthony-chaudhary/fak; repo has .claude, .codex, AGENTS.md. View source.
Similar vibe-coded apps
Shared persistent memory for AI coding tools across sessions and applications
Local context engine that prepares task-focused file packs for AI coding agents
Open source AI gateway routing across 350+ providers through one endpoint
Open-source AI job search agent that runs locally in your CLI
Local code graph for AI-assisted reviews with 63x context reduction
Open-source framework for building autonomous agents with purpose-built UIs
Prescriptive methodology and agent skills for spec-driven development with AI